[kaffe] DateFormatTest regression

[Ito Kazumitsu]

In message "Re: [kaffe] DateFormatTest regression"
    on 04/03/04, Ito Kazumitsu <ito.kazumitsu at hitachi-cable.co.jp> writes:

> When you give the timezone by TZ, the zoneinfo from OS, which is
> free from DOWIM_BUG and says DST starts on the last Sunday in March,
> is used and DST_BUG does not occur until the last Sunday in March.

This is not correct (old kaffe's TimeZone loaded OS's timezone
files but not any more).

When you give the timezone by TZ, libraries/clib/native/System.c
sets user.timezone in this way:

        tzone = "GMT";
#if defined(HAVE_TM_ZONE) && defined(HAVE_LOCALTIME)
        {
                const time_t now = time(NULL);

                if (now != (time_t) -1) {
                        tzone = (char*)localtime(&now)->tm_zone;
                }
        }
#endif
        setProperty(p, "user.timezone", tzone);

In case of TZ=/usr/share/zoneinfo/America/Scoresbysund,
user.timezone is set to "EGT".  As I see it, "EGT" seemes
to be treated as an unknown time zone,  that is almost
equivalent to "GMT".